Related Experiment Video
Updated: Jun 1, 2026
12:43
A Protocol for Comprehensive Assessment of Bulbar Dysfunction in Amyotrophic Lateral Sclerosis (ALS)
Published on: February 21, 2011
Formant analysis in dysphonic patients and automatic Arabic digit speech recognition
Ghulam Muhammad1, Tamer A Mesallam, Khalid H Malki
1Computer Engineering Department, College of Computer and Information Sciences, King Saud University, Riyadh, Saudi Arabia. ghulam@ksu.edu.sa
Biomedical Engineering Online
|June 1, 2011
Summary
Automatic speech recognition (ASR) systems show perfect accuracy for normal speech but struggle with dysphonic voices. Current ASR technology is unreliable for classifying pathological voice disorders.
Area of Science:
- Speech processing
- Bioacoustics
- Medical informatics
Background:
- Objective assessment of speech in dysphonic patients is gaining interest for voice pathology classification.
- Automatic speech recognition (ASR) systems are being explored for this purpose.
Purpose of the Study:
- To evaluate the accuracy of a conventional ASR system in recognizing speech characteristics of individuals with pathological voices.
- To assess the effectiveness of Mel frequency cepstral coefficients (MFCCs) and hidden Markov models (HMMs) in this context.
Main Methods:
- Analyzed speech samples from 62 dysphonic patients and 50 normal subjects speaking Arabic digits.
- Extracted formant distributions of the vowel /a/ to identify deviations from normal patterns.
Main Results:
- Achieved 100% recognition accuracy for normal speakers.
- Observed a significant decrease in accuracy for dysphonic speakers.
- Found no improvement in ASR performance post-treatment for voice-disordered subjects.
Conclusions:
- Conventional ASR systems, utilizing MFCCs and HMMs, are not reliable for recognizing speech in dysphonic patients.
- The current ASR technique is inadequate for classifying voice pathologies.
